From 73ced1f90ebfcd04009929b831f0d7f93eab1f93 Mon Sep 17 00:00:00 2001 From: Dmitry Shilov Date: Tue, 18 Aug 2026 08:43:36 +0300 Subject: [PATCH 1/3] Add Arka-AI provider (OpenAI-compatible model hub) Adds Arka-AI (https://arka-ai.ru), an OpenAI-compatible model hub at https://api.arka-ai.ru/v1, with 23 chat/reasoning models sourced from DeepSeek, Qwen (Alibaba), Kimi (Moonshot), MiniMax, GLM (Zhipu), and Step. Provider files are override-only with base_model pointing at existing lab entries. reasoning_options are copied from each underlying lab's first-party entry (toggle / effort / budget_tokens as applicable). Cost is converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. Logo is the official Arka AI mark (currentColor). --- providers/arka-ai/logo.svg | 6 ++++++ providers/arka-ai/models/DeepSeek-R1.toml | 12 ++++++++++++ providers/arka-ai/models/DeepSeek-V3.toml | 8 ++++++++ providers/arka-ai/models/DeepSeek-V4-Pro.toml | 15 +++++++++++++++ providers/arka-ai/models/GLM-4.7-Flash.toml | 12 ++++++++++++ providers/arka-ai/models/GLM-5.1.toml | 14 ++++++++++++++ providers/arka-ai/models/GLM-5.2.toml | 14 ++++++++++++++ providers/arka-ai/models/GLM-5.toml | 14 ++++++++++++++ providers/arka-ai/models/Kimi-K2.7-Code.toml | 12 ++++++++++++ providers/arka-ai/models/MiniMax-M2.5.toml | 9 +++++++++ providers/arka-ai/models/MiniMax-M3.toml | 11 +++++++++++ providers/arka-ai/models/Qwen3-Coder-Flash.toml | 8 ++++++++ providers/arka-ai/models/Qwen3-Coder-Next.toml | 8 ++++++++ providers/arka-ai/models/Qwen3.5-27B.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.5-Flash.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.5-Plus.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.6-Max.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.6-Plus.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.7-Max.toml | 12 ++++++++++++ providers/arka-ai/models/Qwen3.7-Plus.toml | 12 ++++++++++++ providers/arka-ai/models/Step-3.7-Flash.toml | 14 ++++++++++++++ .../arka-ai/models/deepseek-v4-flash-0731.toml | 15 +++++++++++++++ providers/arka-ai/models/kimi-k3.toml | 15 +++++++++++++++ providers/arka-ai/models/qwen3.8-max.toml | 16 ++++++++++++++++ providers/arka-ai/provider.toml | 5 +++++ 25 files changed, 292 insertions(+) create mode 100644 providers/arka-ai/logo.svg create mode 100644 providers/arka-ai/models/DeepSeek-R1.toml create mode 100644 providers/arka-ai/models/DeepSeek-V3.toml create mode 100644 providers/arka-ai/models/DeepSeek-V4-Pro.toml create mode 100644 providers/arka-ai/models/GLM-4.7-Flash.toml create mode 100644 providers/arka-ai/models/GLM-5.1.toml create mode 100644 providers/arka-ai/models/GLM-5.2.toml create mode 100644 providers/arka-ai/models/GLM-5.toml create mode 100644 providers/arka-ai/models/Kimi-K2.7-Code.toml create mode 100644 providers/arka-ai/models/MiniMax-M2.5.toml create mode 100644 providers/arka-ai/models/MiniMax-M3.toml create mode 100644 providers/arka-ai/models/Qwen3-Coder-Flash.toml create mode 100644 providers/arka-ai/models/Qwen3-Coder-Next.toml create mode 100644 providers/arka-ai/models/Qwen3.5-27B.toml create mode 100644 providers/arka-ai/models/Qwen3.5-Flash.toml create mode 100644 providers/arka-ai/models/Qwen3.5-Plus.toml create mode 100644 providers/arka-ai/models/Qwen3.6-Max.toml create mode 100644 providers/arka-ai/models/Qwen3.6-Plus.toml create mode 100644 providers/arka-ai/models/Qwen3.7-Max.toml create mode 100644 providers/arka-ai/models/Qwen3.7-Plus.toml create mode 100644 providers/arka-ai/models/Step-3.7-Flash.toml create mode 100644 providers/arka-ai/models/deepseek-v4-flash-0731.toml create mode 100644 providers/arka-ai/models/kimi-k3.toml create mode 100644 providers/arka-ai/models/qwen3.8-max.toml create mode 100644 providers/arka-ai/provider.toml diff --git a/providers/arka-ai/logo.svg b/providers/arka-ai/logo.svg new file mode 100644 index 0000000000..2c722ab03b --- /dev/null +++ b/providers/arka-ai/logo.svg @@ -0,0 +1,6 @@ + + + + + + diff --git a/providers/arka-ai/models/DeepSeek-R1.toml b/providers/arka-ai/models/DeepSeek-R1.toml new file mode 100644 index 0000000000..e58fec5ca1 --- /dev/null +++ b/providers/arka-ai/models/DeepSeek-R1.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "deepseek/deepseek-r1" +reasoning_options = [] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.712923 +output = 2.851693 diff --git a/providers/arka-ai/models/DeepSeek-V3.toml b/providers/arka-ai/models/DeepSeek-V3.toml new file mode 100644 index 0000000000..40b0f5a144 --- /dev/null +++ b/providers/arka-ai/models/DeepSeek-V3.toml @@ -0,0 +1,8 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "deepseek/deepseek-v3" + +[cost] +input = 0.356462 +output = 1.425847 diff --git a/providers/arka-ai/models/DeepSeek-V4-Pro.toml b/providers/arka-ai/models/DeepSeek-V4-Pro.toml new file mode 100644 index 0000000000..e7cce02de8 --- /dev/null +++ b/providers/arka-ai/models/DeepSeek-V4-Pro.toml @@ -0,0 +1,15 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "deepseek/deepseek-v4-pro" +reasoning_options = [ + { type = "toggle" }, + { type = "effort", values = ["high", "max"] }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 2.214748 +output = 4.429496 diff --git a/providers/arka-ai/models/GLM-4.7-Flash.toml b/providers/arka-ai/models/GLM-4.7-Flash.toml new file mode 100644 index 0000000000..ee01483294 --- /dev/null +++ b/providers/arka-ai/models/GLM-4.7-Flash.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "zhipuai/glm-4.7-flash" +reasoning_options = [ + { type = "toggle" }, +] + +[interleaved] +field = "reasoning_content" + +# No per-token price published for this model. diff --git a/providers/arka-ai/models/GLM-5.1.toml b/providers/arka-ai/models/GLM-5.1.toml new file mode 100644 index 0000000000..c024ad00ec --- /dev/null +++ b/providers/arka-ai/models/GLM-5.1.toml @@ -0,0 +1,14 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "zhipuai/glm-5.1" +reasoning_options = [ + { type = "toggle" }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.425847 +output = 4.990464 diff --git a/providers/arka-ai/models/GLM-5.2.toml b/providers/arka-ai/models/GLM-5.2.toml new file mode 100644 index 0000000000..3362a52914 --- /dev/null +++ b/providers/arka-ai/models/GLM-5.2.toml @@ -0,0 +1,14 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "zhipuai/glm-5.2" +reasoning_options = [ + { type = "effort", values = ["high", "max"] }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.425847 +output = 4.990464 diff --git a/providers/arka-ai/models/GLM-5.toml b/providers/arka-ai/models/GLM-5.toml new file mode 100644 index 0000000000..a696d573dc --- /dev/null +++ b/providers/arka-ai/models/GLM-5.toml @@ -0,0 +1,14 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "zhipuai/glm-5" +reasoning_options = [ + { type = "toggle" }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.712923 +output = 3.208155 diff --git a/providers/arka-ai/models/Kimi-K2.7-Code.toml b/providers/arka-ai/models/Kimi-K2.7-Code.toml new file mode 100644 index 0000000000..a68c44628a --- /dev/null +++ b/providers/arka-ai/models/Kimi-K2.7-Code.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "moonshotai/kimi-k2.7-code" +reasoning_options = [] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 1.1585 +output = 4.812233 diff --git a/providers/arka-ai/models/MiniMax-M2.5.toml b/providers/arka-ai/models/MiniMax-M2.5.toml new file mode 100644 index 0000000000..9c3e1eb23b --- /dev/null +++ b/providers/arka-ai/models/MiniMax-M2.5.toml @@ -0,0 +1,9 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "minimax/MiniMax-M2.5" +reasoning_options = [] + +[cost] +input = 0.374285 +output = 1.497139 diff --git a/providers/arka-ai/models/MiniMax-M3.toml b/providers/arka-ai/models/MiniMax-M3.toml new file mode 100644 index 0000000000..c679f7d59e --- /dev/null +++ b/providers/arka-ai/models/MiniMax-M3.toml @@ -0,0 +1,11 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "minimax/MiniMax-M3" +reasoning_options = [ + { type = "toggle" }, +] + +[cost] +input = 0.374285 +output = 1.497139 diff --git a/providers/arka-ai/models/Qwen3-Coder-Flash.toml b/providers/arka-ai/models/Qwen3-Coder-Flash.toml new file mode 100644 index 0000000000..b2c6c70a32 --- /dev/null +++ b/providers/arka-ai/models/Qwen3-Coder-Flash.toml @@ -0,0 +1,8 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3-coder-flash" + +[cost] +input = 0.178231 +output = 0.712923 diff --git a/providers/arka-ai/models/Qwen3-Coder-Next.toml b/providers/arka-ai/models/Qwen3-Coder-Next.toml new file mode 100644 index 0000000000..93c1ea6d04 --- /dev/null +++ b/providers/arka-ai/models/Qwen3-Coder-Next.toml @@ -0,0 +1,8 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3-coder-next" + +[cost] +input = 0.712923 +output = 0.712923 diff --git a/providers/arka-ai/models/Qwen3.5-27B.toml b/providers/arka-ai/models/Qwen3.5-27B.toml new file mode 100644 index 0000000000..26bb1a895e --- /dev/null +++ b/providers/arka-ai/models/Qwen3.5-27B.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.5-27b" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 0.106939 +output = 0.855508 diff --git a/providers/arka-ai/models/Qwen3.5-Flash.toml b/providers/arka-ai/models/Qwen3.5-Flash.toml new file mode 100644 index 0000000000..cc4b62a9ee --- /dev/null +++ b/providers/arka-ai/models/Qwen3.5-Flash.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.5-flash" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 0.035646 +output = 0.356462 diff --git a/providers/arka-ai/models/Qwen3.5-Plus.toml b/providers/arka-ai/models/Qwen3.5-Plus.toml new file mode 100644 index 0000000000..edf68aac84 --- /dev/null +++ b/providers/arka-ai/models/Qwen3.5-Plus.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.5-plus" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 0.142585 +output = 0.855508 diff --git a/providers/arka-ai/models/Qwen3.6-Max.toml b/providers/arka-ai/models/Qwen3.6-Max.toml new file mode 100644 index 0000000000..2735b91413 --- /dev/null +++ b/providers/arka-ai/models/Qwen3.6-Max.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.6-max-preview" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 1.604078 +output = 9.624465 diff --git a/providers/arka-ai/models/Qwen3.6-Plus.toml b/providers/arka-ai/models/Qwen3.6-Plus.toml new file mode 100644 index 0000000000..2311ec5af8 --- /dev/null +++ b/providers/arka-ai/models/Qwen3.6-Plus.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.6-plus" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 1.425847 +output = 8.55508 diff --git a/providers/arka-ai/models/Qwen3.7-Max.toml b/providers/arka-ai/models/Qwen3.7-Max.toml new file mode 100644 index 0000000000..be8b58b99f --- /dev/null +++ b/providers/arka-ai/models/Qwen3.7-Max.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.7-max" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 2.13877 +output = 6.41631 diff --git a/providers/arka-ai/models/Qwen3.7-Plus.toml b/providers/arka-ai/models/Qwen3.7-Plus.toml new file mode 100644 index 0000000000..6a4fba2f24 --- /dev/null +++ b/providers/arka-ai/models/Qwen3.7-Plus.toml @@ -0,0 +1,12 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.7-plus" +reasoning_options = [ + { type = "toggle" }, + { type = "budget_tokens" }, +] + +[cost] +input = 0.356462 +output = 1.425847 diff --git a/providers/arka-ai/models/Step-3.7-Flash.toml b/providers/arka-ai/models/Step-3.7-Flash.toml new file mode 100644 index 0000000000..f50e4aad43 --- /dev/null +++ b/providers/arka-ai/models/Step-3.7-Flash.toml @@ -0,0 +1,14 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "stepfun/step-3.7-flash" +reasoning_options = [ + { type = "effort", values = ["low", "medium", "high"] }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.240612 +output = 1.44367 diff --git a/providers/arka-ai/models/deepseek-v4-flash-0731.toml b/providers/arka-ai/models/deepseek-v4-flash-0731.toml new file mode 100644 index 0000000000..9bca4a6527 --- /dev/null +++ b/providers/arka-ai/models/deepseek-v4-flash-0731.toml @@ -0,0 +1,15 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "deepseek/deepseek-v4-flash-0731" +reasoning_options = [ + { type = "toggle" }, + { type = "effort", values = ["high", "max"] }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.178231 +output = 0.356462 diff --git a/providers/arka-ai/models/kimi-k3.toml b/providers/arka-ai/models/kimi-k3.toml new file mode 100644 index 0000000000..075464825d --- /dev/null +++ b/providers/arka-ai/models/kimi-k3.toml @@ -0,0 +1,15 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "moonshotai/kimi-k3" +reasoning_options = [ + { type = "toggle" }, + { type = "effort", values = ["low", "high", "max"] }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 3.564617 +output = 17.823084 diff --git a/providers/arka-ai/models/qwen3.8-max.toml b/providers/arka-ai/models/qwen3.8-max.toml new file mode 100644 index 0000000000..44a41b6de8 --- /dev/null +++ b/providers/arka-ai/models/qwen3.8-max.toml @@ -0,0 +1,16 @@ +# Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). +# Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. + +base_model = "alibaba/qwen3.8-max" +reasoning_options = [ + { type = "toggle" }, + { type = "effort", values = ["low", "medium", "xhigh"] }, + { type = "budget_tokens", min = 0, max = 262144 }, +] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 2.13877 +output = 6.41631 diff --git a/providers/arka-ai/provider.toml b/providers/arka-ai/provider.toml new file mode 100644 index 0000000000..f6f180f0ae --- /dev/null +++ b/providers/arka-ai/provider.toml @@ -0,0 +1,5 @@ +name = "Arka-AI" +npm = "@ai-sdk/openai-compatible" +env = ["ARKA_API_KEY"] +api = "https://api.arka-ai.ru/v1" +doc = "https://arka-ai.ru" From 700137002a527fe10d1feaeb2b3904c4f2022bc4 Mon Sep 17 00:00:00 2001 From: Dmitry Shilov Date: Tue, 18 Aug 2026 09:40:03 +0300 Subject: [PATCH 2/3] fix(arka-ai): document toggle wire paths, align effort levels, drop unenforced controls - Add top-of-file wire-path comments to all 15 toggle model files, verified live against https://api.arka-ai.ru/v1/chat/completions: enable_thinking true|false is the canonical toggle (thinking.type = enabled|disabled also accepted), except Qwen3.5-27B where only enable_thinking takes effect. - deepseek-v4-flash-0731: effort levels low|high|max all verified on Arka, matching the first-party deepseek baseline. - kimi-k3: reasoning is always on; disabling it 400s on the non-thinking lane and effort controls have no observable effect, so reasoning_options is now []. - GLM-4.7-Flash: keep toggle, evidenced from the zhipuai lab baseline and Arka's verified GLM siblings (route returned upstream_error at verification time). - Remove budget_tokens from 8 Qwen files: thinking_budget is accepted but not enforced (reasoning length unchanged at budget=64..32768, streaming and non-streaming). --- providers/arka-ai/models/DeepSeek-V4-Pro.toml | 3 +++ providers/arka-ai/models/GLM-4.7-Flash.toml | 5 +++++ providers/arka-ai/models/GLM-5.1.toml | 3 +++ providers/arka-ai/models/GLM-5.toml | 3 +++ providers/arka-ai/models/MiniMax-M3.toml | 4 ++++ providers/arka-ai/models/Qwen3.5-27B.toml | 7 ++++++- providers/arka-ai/models/Qwen3.5-Flash.toml | 5 ++++- providers/arka-ai/models/Qwen3.5-Plus.toml | 5 ++++- providers/arka-ai/models/Qwen3.6-Max.toml | 5 ++++- providers/arka-ai/models/Qwen3.6-Plus.toml | 5 ++++- providers/arka-ai/models/Qwen3.7-Max.toml | 5 ++++- providers/arka-ai/models/Qwen3.7-Plus.toml | 5 ++++- providers/arka-ai/models/deepseek-v4-flash-0731.toml | 6 +++++- providers/arka-ai/models/kimi-k3.toml | 10 ++++++---- providers/arka-ai/models/qwen3.8-max.toml | 6 +++++- 15 files changed, 64 insertions(+), 13 deletions(-) diff --git a/providers/arka-ai/models/DeepSeek-V4-Pro.toml b/providers/arka-ai/models/DeepSeek-V4-Pro.toml index e7cce02de8..79f4c00688 100644 --- a/providers/arka-ai/models/DeepSeek-V4-Pro.toml +++ b/providers/arka-ai/models/DeepSeek-V4-Pro.toml @@ -1,3 +1,6 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Effort: reasoning_effort = high|max (low accepted; maps per DeepSeek lab convention) +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. diff --git a/providers/arka-ai/models/GLM-4.7-Flash.toml b/providers/arka-ai/models/GLM-4.7-Flash.toml index ee01483294..dc26f14685 100644 --- a/providers/arka-ai/models/GLM-4.7-Flash.toml +++ b/providers/arka-ai/models/GLM-4.7-Flash.toml @@ -1,3 +1,8 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled, +# the zhipuai-native wire). First-party providers/zhipuai/models/glm-4.7-flash.toml +# declares the same toggle. This route returned upstream_error (HTTP 400) at verification +# time, so the wire is evidenced from Arka's verified GLM siblings (GLM-5, GLM-5.1) +# served through the same gateway (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. diff --git a/providers/arka-ai/models/GLM-5.1.toml b/providers/arka-ai/models/GLM-5.1.toml index c024ad00ec..f0756439b1 100644 --- a/providers/arka-ai/models/GLM-5.1.toml +++ b/providers/arka-ai/models/GLM-5.1.toml @@ -1,3 +1,6 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled, +# the zhipuai-native wire) +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. diff --git a/providers/arka-ai/models/GLM-5.toml b/providers/arka-ai/models/GLM-5.toml index a696d573dc..04d4427030 100644 --- a/providers/arka-ai/models/GLM-5.toml +++ b/providers/arka-ai/models/GLM-5.toml @@ -1,3 +1,6 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled, +# the zhipuai-native wire) +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. diff --git a/providers/arka-ai/models/MiniMax-M3.toml b/providers/arka-ai/models/MiniMax-M3.toml index c679f7d59e..13fa8735ce 100644 --- a/providers/arka-ai/models/MiniMax-M3.toml +++ b/providers/arka-ai/models/MiniMax-M3.toml @@ -1,3 +1,7 @@ +# Toggle: enable_thinking true|false (thinking.type = disabled also accepted). +# reasoning_content is not exposed; reasoning is visible only via +# usage.completion_tokens_details.reasoning_tokens. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. diff --git a/providers/arka-ai/models/Qwen3.5-27B.toml b/providers/arka-ai/models/Qwen3.5-27B.toml index 26bb1a895e..ba72e3b164 100644 --- a/providers/arka-ai/models/Qwen3.5-27B.toml +++ b/providers/arka-ai/models/Qwen3.5-27B.toml @@ -1,10 +1,15 @@ +# Toggle: enable_thinking true|false only; thinking.type = disabled is accepted but has no +# effect on this model. Reasoning is not exposed separately (no reasoning_content or +# reasoning_tokens); the toggle is verified via completion length. +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.5-27b" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.5-Flash.toml b/providers/arka-ai/models/Qwen3.5-Flash.toml index cc4b62a9ee..d313c2fa33 100644 --- a/providers/arka-ai/models/Qwen3.5-Flash.toml +++ b/providers/arka-ai/models/Qwen3.5-Flash.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced (reasoning length unchanged at +# budget=256 vs no budget), so no budget_tokens control is advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.5-flash" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.5-Plus.toml b/providers/arka-ai/models/Qwen3.5-Plus.toml index edf68aac84..e819f62c05 100644 --- a/providers/arka-ai/models/Qwen3.5-Plus.toml +++ b/providers/arka-ai/models/Qwen3.5-Plus.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.5-plus" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.6-Max.toml b/providers/arka-ai/models/Qwen3.6-Max.toml index 2735b91413..18cd22e272 100644 --- a/providers/arka-ai/models/Qwen3.6-Max.toml +++ b/providers/arka-ai/models/Qwen3.6-Max.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.6-max-preview" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.6-Plus.toml b/providers/arka-ai/models/Qwen3.6-Plus.toml index 2311ec5af8..51a862bda1 100644 --- a/providers/arka-ai/models/Qwen3.6-Plus.toml +++ b/providers/arka-ai/models/Qwen3.6-Plus.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.6-plus" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.7-Max.toml b/providers/arka-ai/models/Qwen3.7-Max.toml index be8b58b99f..27b292ea50 100644 --- a/providers/arka-ai/models/Qwen3.7-Max.toml +++ b/providers/arka-ai/models/Qwen3.7-Max.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.7-max" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/Qwen3.7-Plus.toml b/providers/arka-ai/models/Qwen3.7-Plus.toml index 6a4fba2f24..7fd4249131 100644 --- a/providers/arka-ai/models/Qwen3.7-Plus.toml +++ b/providers/arka-ai/models/Qwen3.7-Plus.toml @@ -1,10 +1,13 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Budget: thinking_budget is accepted but not enforced, so no budget_tokens control is +# advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "alibaba/qwen3.7-plus" reasoning_options = [ { type = "toggle" }, - { type = "budget_tokens" }, ] [cost] diff --git a/providers/arka-ai/models/deepseek-v4-flash-0731.toml b/providers/arka-ai/models/deepseek-v4-flash-0731.toml index 9bca4a6527..35533a68f0 100644 --- a/providers/arka-ai/models/deepseek-v4-flash-0731.toml +++ b/providers/arka-ai/models/deepseek-v4-flash-0731.toml @@ -1,10 +1,14 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Effort: reasoning_effort = low|high|max (all three levels verified live, matching the +# first-party deepseek baseline; unlike alibaba's hosted copy, low is not mapped away) +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "deepseek/deepseek-v4-flash-0731" reasoning_options = [ { type = "toggle" }, - { type = "effort", values = ["high", "max"] }, + { type = "effort", values = ["low", "high", "max"] }, ] [interleaved] diff --git a/providers/arka-ai/models/kimi-k3.toml b/providers/arka-ai/models/kimi-k3.toml index 075464825d..3310377d43 100644 --- a/providers/arka-ai/models/kimi-k3.toml +++ b/providers/arka-ai/models/kimi-k3.toml @@ -1,11 +1,13 @@ +# Reasoning is always on: disabling it (enable_thinking=false or thinking.type=disabled) +# returns HTTP 400 "invalid temperature: only 0.6 is allowed" from the non-thinking lane, +# and effort controls (reasoning_effort, output_config.effort) are accepted without any +# observable change in reasoning depth. No client control is therefore advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. base_model = "moonshotai/kimi-k3" -reasoning_options = [ - { type = "toggle" }, - { type = "effort", values = ["low", "high", "max"] }, -] +reasoning_options = [] [interleaved] field = "reasoning_content" diff --git a/providers/arka-ai/models/qwen3.8-max.toml b/providers/arka-ai/models/qwen3.8-max.toml index 44a41b6de8..05659630a3 100644 --- a/providers/arka-ai/models/qwen3.8-max.toml +++ b/providers/arka-ai/models/qwen3.8-max.toml @@ -1,3 +1,8 @@ +# Toggle: enable_thinking true|false (Arka also accepts thinking.type = enabled|disabled) +# Effort: reasoning_effort = low|medium|xhigh (high and max accepted as aliases) +# Budget: thinking_budget is accepted but not enforced (reasoning length unchanged at +# budget=64 vs budget=16384), so no budget_tokens control is advertised. +# Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models. @@ -5,7 +10,6 @@ base_model = "alibaba/qwen3.8-max" reasoning_options = [ { type = "toggle" }, { type = "effort", values = ["low", "medium", "xhigh"] }, - { type = "budget_tokens", min = 0, max = 262144 }, ] [interleaved] From 92ea8e99fd1e1330f5dad6ca200a5d20bc64c24d Mon Sep 17 00:00:00 2001 From: Dmitry Shilov Date: Tue, 18 Aug 2026 09:46:24 +0300 Subject: [PATCH 3/3] fix(arka-ai): clarify kimi-k3 non-thinking lane is unusable even at temperature=0.6 --- providers/arka-ai/models/kimi-k3.toml | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/providers/arka-ai/models/kimi-k3.toml b/providers/arka-ai/models/kimi-k3.toml index 3310377d43..1940d9ebae 100644 --- a/providers/arka-ai/models/kimi-k3.toml +++ b/providers/arka-ai/models/kimi-k3.toml @@ -1,7 +1,9 @@ # Reasoning is always on: disabling it (enable_thinking=false or thinking.type=disabled) -# returns HTTP 400 "invalid temperature: only 0.6 is allowed" from the non-thinking lane, -# and effort controls (reasoning_effort, output_config.effort) are accepted without any -# observable change in reasoning depth. No client control is therefore advertised. +# returns HTTP 400 "invalid temperature: only 0.6 is allowed" from the non-thinking lane +# even when temperature=0.6 is explicitly set (the value the error demands), so that lane +# is unusable by callers. Effort controls (reasoning_effort, output_config.effort, +# thinking.type=adaptive + output_config.effort) are accepted without any observable +# change in reasoning depth. No client control is therefore advertised. # Verified live on https://api.arka-ai.ru/v1/chat/completions (accessed 2026-08-18). # Cost: converted from Arka AI's RUB list price at 82.9977 RUB/USD (CBR rate captured 2026-08-13). # Sources: https://arka-ai.ru (pricing) and https://api.arka-ai.ru/v1/models.